Re: Amazing work on bug fixing
Gunter K�nigsmann via Maxima-discuss <[email protected]> Mon, 22 Jun 2026 07:30:26 +0200
| Newsgroups | gmane.comp.mathematics.maxima.general |
|---|---|
| Message-ID | <[email protected]> |
--===============1126443113243486631==
Content-Type: multipart/alternative;
boundary=----L3ZBZCIORG9H5LFQ4T89EJX623RYKX
Content-Transfer-Encoding: 7bit
------L3ZBZCIORG9H5LFQ4T89EJX623RYKX
Content-Type: text/plain;
charset=utf-8
Content-Transfer-Encoding: quoted-printable
With Gemini I learned that what it can do from a human view can be very une=
ven:=20
It can fulfill big tasks ("create a diff viewer") I'm seconds=2E But in ot=
her cases it might be unable to pull off an obvious (at least for a human b=
eing) 2-line fix correctly even if you describe it what to do in detail=2E=
=20
It was obviously trained never to give up even if it doesn't have a clue o=
n what goes on and it if will hit a fix that will be by accident: If Gemini=
starts bragging on how good and elegant this fix was that is a typical sig=
n that it hopes you will believe that and won't verify the code=2E Also it=
won't speak up if what you ask it to doesn't look like it would work=2E
And it is quite good in doing complicated stuff like complex RegEx work, b=
ut trusts its abilities=2E=20
Claude on the other hand doesn't trust its abilities, but delegates much w=
ork to scripts it writes for that purpose=2E Then it first tests the script=
s, provides them with debug output that I'm detail explains what assumption=
s in the end failed and what to do in that case=2E Then it looks at their r=
esults and then it writes a formal test for the results=2E When a Mythos-cl=
ass AI was available for a few days I learned that this one spoke up if you=
went into a direction that it didn't understand=2E
Don't know how big Claude's grasp is on mathematics and if Claude's limite=
d context window is big enough to wrap it's head about complex Maxima Probl=
ems=2E But for C++ at least that approach normally provides better results=
=2E Also Claude does do what you do by hand for Gemini: It organises it's w=
ork by splitting it into bite-sized tasks that each leads a step into the R=
ight Direction=2E=20
Also I seem to have noticed that AIs might act like a overworked intern du=
ring the US 9 to five hours while doing excellent work if no-one else assig=
ns it tasks to=2E=20
Kind regards,=20
Gunter
On 21 June 2026 22:56:42 CEST, ptorpey00@gmail=2Ecom wrote:
>In case you get mostly incorrect fixes:
>
>* Don't continue the conversation for too long, as it will eventually for=
get the first part of it, which contains the highly important source code=
=2E LLMs have a limited context window=2E If the conversation gets too long=
, it starts forgetting things said earlier=2E
>* Instead, edit your initial prompt and include a section like: "You prev=
iously proposed the following fixes, which didn't work: [=2E=2E=2E]"=2E Inc=
lude a short description for every proposed fix and why it wasn't working (=
e=2Eg=2E it broke certain tests)=2E If editing the initial prompt is not po=
ssible because you have sent follow-up prompts already, you can "fork" the =
conversation after the first reply, and then you will be able to edit the i=
nitial prompt in the forked conversation=2E
>
>[PT] Good point=2E If the context window gets too long, the LLM will go =
off the rails=2E Another similar approach I=E2=80=99ve used to continue t=
he conversation is to tell the LLM in the original session that you are goi=
ng to start over in a new chat session and to write a prompt for you that w=
ill provide background info on the task to be solved, what worked successfu=
lly thus far, what assumptions the LLM might have gotten wrong in the conve=
rsation, and any other info that would be useful to continue the task=2E I=
often use the LLM to write prompts like this when starting a big task=2E =
It is also good to ask the LLM if it overlooked anything or has any additio=
nal suggestions for creating the prompt=2E Amazing what these tools can do=
these days, and they are improving very fast=2E So if something didn=E2=
=80=99t work yesterday, try again today!
>
>=20
>
>--Pete
>
>=20
>
------L3ZBZCIORG9H5LFQ4T89EJX623RYKX
Content-Type: text/html;
charset=utf-8
Content-Transfer-Encoding: quoted-printable
<html xmlns:v=3D"urn:schemas-microsoft-com:vml" xmlns:o=3D"urn:schemas-micr=
osoft-com:office:office" xmlns:w=3D"urn:schemas-microsoft-com:office:word" =
xmlns:m=3D"http://schemas=2Emicrosoft=2Ecom/office/2004/12/omml" xmlns=3D"h=
ttp://www=2Ew3=2Eorg/TR/REC-html40"><head><meta http-equiv=3D"Content-Type"=
content=3D"text/html; charset=3Dutf-8"><meta name=3D"Generator" content=3D=
"Microsoft Word 15 (filtered medium)"><style><!--
/* Font Definitions */
@font-face
{font-family:Wingdings;
panose-1:5 0 0 0 0 0 0 0 0 0;}
@font-face
{font-family:"Cambria Math";
panose-1:2 4 5 3 5 4 6 3 2 4;}
@font-face
{font-family:Aptos;}
/* Style Definitions */
p=2EMsoNormal, li=2EMsoNormal, div=2EMsoNormal
{margin:0in;
font-size:12=2E0pt;
font-family:"Aptos",sans-serif;}
span=2EEmailStyle18
{mso-style-type:personal-reply;
font-family:"Aptos",sans-serif;
color:windowtext;}
=2EMsoChpDefault
{mso-style-type:export-only;}
@page WordSection1
{size:8=2E5in 11=2E0in;
margin:1=2E0in 1=2E0in 1=2E0in 1=2E0in;}
div=2EWordSection1
{page:WordSection1;}
/* List Definitions */
@list l0
{mso-list-id:475877412;
mso-list-template-ids:1386390900;}
@list l0:level1
{mso-level-number-format:bullet;
mso-level-text:=EF=82=B7;
mso-level-tab-stop:=2E5in;
mso-level-number-position:left;
text-indent:-=2E25in;
mso-ansi-font-size:10=2E0pt;
font-family:Symbol;}
@list l0:level2
{mso-level-number-format:bullet;
mso-level-text:o;
mso-level-tab-stop:1=2E0in;
mso-level-number-position:left;
text-indent:-=2E25in;
mso-ansi-font-size:10=2E0pt;
font-family:"Courier New";
mso-bidi-font-family:"Times New Roman";}
@list l0:level3
{mso-level-number-format:bullet;
mso-level-text:=EF=82=A7;
mso-level-tab-stop:1=2E5in;
mso-level-number-position:left;
text-indent:-=2E25in;
mso-ansi-font-size:10=2E0pt;
font-family:Wingdings;}
@list l0:level4
{mso-level-number-format:bullet;
mso-level-text:=EF=82=A7;
mso-level-tab-stop:2=2E0in;
mso-level-number-position:left;
text-indent:-=2E25in;
mso-ansi-font-size:10=2E0pt;
font-family:Wingdings;}
@list l0:level5
{mso-level-number-format:bullet;
mso-level-text:=EF=82=A7;
mso-level-tab-stop:2=2E5in;
mso-level-number-position:left;
text-indent:-=2E25in;
mso-ansi-font-size:10=2E0pt;
font-family:Wingdings;}
@list l0:level6
{mso-level-number-format:bullet;
mso-level-text:=EF=82=A7;
mso-level-tab-stop:3=2E0in;
mso-level-number-position:left;
text-indent:-=2E25in;
mso-ansi-font-size:10=2E0pt;
font-family:Wingdings;}
@list l0:level7
{mso-level-number-format:bullet;
mso-level-text:=EF=82=A7;
mso-level-tab-stop:3=2E5in;
mso-level-number-position:left;
text-indent:-=2E25in;
mso-ansi-font-size:10=2E0pt;
font-family:Wingdings;}
@list l0:level8
{mso-level-number-format:bullet;
mso-level-text:=EF=82=A7;
mso-level-tab-stop:4=2E0in;
mso-level-number-position:left;
text-indent:-=2E25in;
mso-ansi-font-size:10=2E0pt;
font-family:Wingdings;}
@list l0:level9
{mso-level-number-format:bullet;
mso-level-text:=EF=82=A7;
mso-level-tab-stop:4=2E5in;
mso-level-number-position:left;
text-indent:-=2E25in;
mso-ansi-font-size:10=2E0pt;
font-family:Wingdings;}
@list l1
{mso-list-id:841047471;
mso-list-template-ids:-1627226824;}
@list l1:level1
{mso-level-number-format:bullet;
mso-level-text:=EF=82=B7;
mso-level-tab-stop:=2E5in;
mso-level-number-position:left;
text-indent:-=2E25in;
mso-ansi-font-size:10=2E0pt;
font-family:Symbol;}
@list l1:level2
{mso-level-number-format:bullet;
mso-level-text:o;
mso-level-tab-stop:1=2E0in;
mso-level-number-position:left;
text-indent:-=2E25in;
mso-ansi-font-size:10=2E0pt;
font-family:"Courier New";
mso-bidi-font-family:"Times New Roman";}
@list l1:level3
{mso-level-number-format:bullet;
mso-level-text:=EF=82=A7;
mso-level-tab-stop:1=2E5in;
mso-level-number-position:left;
text-indent:-=2E25in;
mso-ansi-font-size:10=2E0pt;
font-family:Wingdings;}
@list l1:level4
{mso-level-number-format:bullet;
mso-level-text:=EF=82=A7;
mso-level-tab-stop:2=2E0in;
mso-level-number-position:left;
text-indent:-=2E25in;
mso-ansi-font-size:10=2E0pt;
font-family:Wingdings;}
@list l1:level5
{mso-level-number-format:bullet;
mso-level-text:=EF=82=A7;
mso-level-tab-stop:2=2E5in;
mso-level-number-position:left;
text-indent:-=2E25in;
mso-ansi-font-size:10=2E0pt;
font-family:Wingdings;}
@list l1:level6
{mso-level-number-format:bullet;
mso-level-text:=EF=82=A7;
mso-level-tab-stop:3=2E0in;
mso-level-number-position:left;
text-indent:-=2E25in;
mso-ansi-font-size:10=2E0pt;
font-family:Wingdings;}
@list l1:level7
{mso-level-number-format:bullet;
mso-level-text:=EF=82=A7;
mso-level-tab-stop:3=2E5in;
mso-level-number-position:left;
text-indent:-=2E25in;
mso-ansi-font-size:10=2E0pt;
font-family:Wingdings;}
@list l1:level8
{mso-level-number-format:bullet;
mso-level-text:=EF=82=A7;
mso-level-tab-stop:4=2E0in;
mso-level-number-position:left;
text-indent:-=2E25in;
mso-ansi-font-size:10=2E0pt;
font-family:Wingdings;}
@list l1:level9
{mso-level-number-format:bullet;
mso-level-text:=EF=82=A7;
mso-level-tab-stop:4=2E5in;
mso-level-number-position:left;
text-indent:-=2E25in;
mso-ansi-font-size:10=2E0pt;
font-family:Wingdings;}
@list l2
{mso-list-id:1777750449;
mso-list-template-ids:1573699962;}
@list l2:level1
{mso-level-number-format:bullet;
mso-level-text:=EF=82=B7;
mso-level-tab-stop:=2E5in;
mso-level-number-position:left;
text-indent:-=2E25in;
mso-ansi-font-size:10=2E0pt;
font-family:Symbol;}
@list l2:level2
{mso-level-number-format:bullet;
mso-level-text:o;
mso-level-tab-stop:1=2E0in;
mso-level-number-position:left;
text-indent:-=2E25in;
mso-ansi-font-size:10=2E0pt;
font-family:"Courier New";
mso-bidi-font-family:"Times New Roman";}
@list l2:level3
{mso-level-number-format:bullet;
mso-level-text:=EF=82=A7;
mso-level-tab-stop:1=2E5in;
mso-level-number-position:left;
text-indent:-=2E25in;
mso-ansi-font-size:10=2E0pt;
font-family:Wingdings;}
@list l2:level4
{mso-level-number-format:bullet;
mso-level-text:=EF=82=A7;
mso-level-tab-stop:2=2E0in;
mso-level-number-position:left;
text-indent:-=2E25in;
mso-ansi-font-size:10=2E0pt;
font-family:Wingdings;}
@list l2:level5
{mso-level-number-format:bullet;
mso-level-text:=EF=82=A7;
mso-level-tab-stop:2=2E5in;
mso-level-number-position:left;
text-indent:-=2E25in;
mso-ansi-font-size:10=2E0pt;
font-family:Wingdings;}
@list l2:level6
{mso-level-number-format:bullet;
mso-level-text:=EF=82=A7;
mso-level-tab-stop:3=2E0in;
mso-level-number-position:left;
text-indent:-=2E25in;
mso-ansi-font-size:10=2E0pt;
font-family:Wingdings;}
@list l2:level7
{mso-level-number-format:bullet;
mso-level-text:=EF=82=A7;
mso-level-tab-stop:3=2E5in;
mso-level-number-position:left;
text-indent:-=2E25in;
mso-ansi-font-size:10=2E0pt;
font-family:Wingdings;}
@list l2:level8
{mso-level-number-format:bullet;
mso-level-text:=EF=82=A7;
mso-level-tab-stop:4=2E0in;
mso-level-number-position:left;
text-indent:-=2E25in;
mso-ansi-font-size:10=2E0pt;
font-family:Wingdings;}
@list l2:level9
{mso-level-number-format:bullet;
mso-level-text:=EF=82=A7;
mso-level-tab-stop:4=2E5in;
mso-level-number-position:left;
text-indent:-=2E25in;
mso-ansi-font-size:10=2E0pt;
font-family:Wingdings;}
ol
{margin-bottom:0in;}
ul
{margin-bottom:0in;}
--></style><!--[if gte mso 9]><xml>
<o:shapedefaults v:ext=3D"edit" spidmax=3D"1026" />
</xml><![endif]--><!--[if gte mso 9]><xml>
<o:shapelayout v:ext=3D"edit">
<o:idmap v:ext=3D"edit" data=3D"1" />
</o:shapelayout></xml><![endif]--></head><body lang=3D"EN-US" link=3D"blue=
" vlink=3D"purple" style=3D"word-wrap:break-word"><div dir=3D"auto">With Ge=
mini I learned that what it can do from a human view can be very uneven: <b=
r><br>It can fulfill big tasks ("create a diff viewer") I'm seconds=2E But =
in other cases it might be unable to pull off an obvious (at least for a hu=
man being) 2-line fix correctly even if you describe it what to do in detai=
l=2E <br><br>It was obviously trained never to give up even if it doesn't h=
ave a clue on what goes on and it if will hit a fix that will be by acciden=
t: If Gemini starts bragging on how good and elegant this fix was that is a=
typical sign that it hopes you will believe that and won't verify the code=
=2E=C2=A0 Also it won't speak up if what you ask it to doesn't look like it=
would work=2E<br><br>And it is quite good in doing complicated stuff like =
complex RegEx work, but trusts its abilities=2E <br><br>Claude on the other=
hand doesn't trust its abilities, but delegates much work to scripts it wr=
ites for that purpose=2E Then it first tests the scripts, provides them wit=
h debug output that I'm detail explains what assumptions in the end failed =
and what to do in that case=2E Then it looks at their results and then it w=
rites a formal test for the results=2E When a Mythos-class AI was available=
for a few days I learned that this one spoke up if you went into a directi=
on that it didn't understand=2E<br><br>Don't know how big Claude's grasp is=
on mathematics and if Claude's limited context window is big enough to wra=
p it's head about complex Maxima Problems=2E But for C++ at least that appr=
oach normally provides better results=2E Also Claude does do what you do by=
hand for Gemini: It organises it's work by splitting it into bite-sized ta=
sks that each leads a step into the Right Direction=2E <br><br>Also I seem =
to have noticed that AIs might act like a overworked intern during the US 9=
to five hours while doing excellent work if no-one else assigns it tasks t=
o=2E <br><br>Kind regards, <br><br>=C2=A0=C2=A0 Gunter<br></div><br><br><di=
v class=3D"gmail_quote"><div dir=3D"auto">On 21 June 2026 22:56:42 CEST, pt=
orpey00@gmail=2Ecom wrote:</div><blockquote class=3D"gmail_quote" style=3D"=
margin: 0pt 0pt 0pt 0=2E8ex; border-left: 1px solid rgb(204, 204, 204); pad=
ding-left: 1ex;">
<div class=3D"WordSection1"><div><div><p class=3D"MsoNormal">In case you g=
et mostly incorrect fixes:<o:p></o:p></p></div><div><ul type=3D"disc"><li c=
lass=3D"MsoNormal" style=3D"mso-margin-top-alt:auto;mso-margin-bottom-alt:a=
uto;mso-list:l0 level1 lfo3"><b>Don't continue the conversation for too lon=
g</b>, as it will eventually forget the first part of it, which contains th=
e highly important source code=2E LLMs have a limited context window=2E If =
the conversation gets too long, it starts forgetting things said earlier=2E=
<o:p></o:p></li><li class=3D"MsoNormal" style=3D"mso-margin-top-alt:auto;ms=
o-margin-bottom-alt:auto;mso-list:l0 level1 lfo3">Instead, edit your initia=
l prompt and include a section like: "You previously proposed the following=
fixes, which didn't work: [=2E=2E=2E]"=2E Include a short description for =
every proposed fix and why it wasn't working (e=2Eg=2E it broke certain tes=
ts)=2E If editing the initial prompt is not possible because you have sent =
follow-up prompts already, you can "fork" the conversation after the first =
reply, and then you will be able to edit the initial prompt in the forked c=
onversation=2E<o:p></o:p></li></ul><p class=3D"MsoNormal" style=3D"mso-marg=
in-top-alt:auto;mso-margin-bottom-alt:auto"><b><i>[PT] Good point=2E =
If the context window gets too long, the LLM will go off the rails=2E =
Another similar approach I=E2=80=99ve used to continue the conversat=
ion is to tell the LLM in the original session that you are going to start =
over in a new chat session and to write a prompt for you that will provide =
background info on the task to be solved, what worked successfully thus far=
, what assumptions the LLM might have gotten wrong in the conversation, and=
any other info that would be useful to continue the task=2E I often =
use the LLM to write prompts like this when starting a big task=2E It=
is also good to ask the LLM if it overlooked anything or has any additiona=
l suggestions for creating the prompt=2E Amazing what these tools can=
do these days, and they are improving very fast=2E So if something d=
idn=E2=80=99t work yesterday, try again today!<o:p></o:p></i></b></p><p cla=
ss=3D"MsoNormal" style=3D"mso-margin-top-alt:auto;mso-margin-bottom-alt:aut=
o"><b><i><o:p> </o:p></i></b></p></div></div><div><blockquote style=3D=
"border:none;border-left:solid #CCCCCC 1=2E0pt;padding:0in 0in 0in 6=2E0pt;=
margin-left:4=2E8pt;margin-right:0in"><div><div><p class=3D"MsoNormal">--Pe=
te<o:p></o:p></p><p class=3D"MsoNormal"><o:p> </o:p></p></div></div></=
blockquote></div></div></blockquote></div></body></html>
------L3ZBZCIORG9H5LFQ4T89EJX623RYKX--
--===============1126443113243486631==
Content-Type: text/plain; charset="us-ascii"
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Content-Disposition: inline
--===============1126443113243486631==
Content-Type: text/plain; charset="us-ascii"
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Content-Disposition: inline
_______________________________________________
Maxima-discuss mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/maxima-discuss
--===============1126443113243486631==--